Journal of Health Economics 10 (1991) 107-142. North-Holland
Cost of innovation
industry*
in the pharmaceutical
Joseph A. DiMasi
Center for
the Study of Drug Development,
Tufts University.
Boston,
MA
02111.
USA
Ronald W. Hansen
William E. Simon Graduate
NY 14627, USA
School of Business Administration,
University
of Rochester,
Rochester,
Henry G. Grabowski
Department
of Economics,
Duke University,
Durham,
NC 27706,
USA
Louis Lasagna
Center for
the Study
of Drug
Developments,
Tujis
University,
Boston,
MA
02111,
USA
Received June 1990, final version received November 1990
The research and development costs of 93 randomly selected new chemical entities (NCEs) were
obtained from a survey of 12 U.S.-owned pharmaceutical firms. These data were used to
estimate the pre-tax average cost of new drug development. The costs of abandoned NCEs were
linked to the-costs of NC& that obtained marketing approval. For base case parameter values,
the estimated out-of-pocket cost per approved NCE is $114 million (1987 dollars). Capitalizing
out-of-pocket costs to the point of marketing approval at a 9% discount rate yielded an average
cost estimate of S231 million (1987 dollars).
1. Introduction
ProduCt innovation in the pharmaceutical industry is risky and timeconsuming, with research and development (R&D) costs representing a high
*We are indebted to Gene Trimble who selected our sample of investigational drugs from a
Center for the Study of Drug Development database and assisted in preparing our questionnaire. We are also grateful to Marianne Ulcickas for assistance with data entry. In addition, we
wish to thank the Bradley Policy Research Center at the William E. Simon Graduate School of
Business Administration of the University of Rochester and Duke University’s Program in
Pharmaceutical and Health Economics for partial support of expense-s incurred by Ronald
Hansen and Henry Grabowski, respectively. Finally, we thank the surveyed firms for providing
proprietary data, and a number of individuals in those firms who graciously gave of their time
when we requested clarification of responses.
0167-6296/91/$03.50 0 1991-Elsevier
Science Publishers B.V. (North-Holland)
108
J.A. DiMasi et al., Cost of innovation in pharmaceutical
industry
proportion
of sales revenues.’ Although other forms of pharmaceutical
innovation exist, new chemical entity (NCE) development is, on the whole,
the most therapeutically
and economically significant.2*3 Typically, the
R&D process for NCEs is spread over many years, and only a small
proportion of these new drugs are eventually approved for marketing.
Empirical analyses of the cost to discover and develop NCEs are
interesting on several counts. First, knowledge of R&D costs is important
for analyzing issues such as the returns on R&D investment.4 Second, the
cost of a new drug has a direct bearing on the organizational structure of
innovation in pharmaceuticals. In this regard, higher real R&D costs have
been cited as one of the main factors underlying the recent trend toward
more mergers and industry consolidation. 5 Third, R&D costs also influence
the pattern of international resource allocation, and hence, international
competitiveness.6 Finally, the cost of R&D has become an important issue
in its own right in recent policy deliberations involving regulatory requirements and the economic performance of the pharmaceutical industry.’ In
this paper we confine our analysis specifically to the study of R&D costs.
‘The Pharmaceutical Manufacturers Association (PMA) Statistical Fact Book (1987) reports
that, for member firms, U.S. pharmaceutical expenditures expressed as a percentage of U.S.
pharmaceutical sales was 15.0% in 1986, up from 10.2gb in 1965.
*An NCE is defined here as a new molecular compound not previously tested in humans.
Excluded are new salts and esters of existing compounds; surgical and diagnostic materials;
certain externally used compounds such as disinfectants, antiperspirants, and sunscreens;
nutritional compounds such as natural forms of vitamins and sweetening agents; and certain
biologic compounds such as vaccines, antigens, antisera, immunoglobulins, or purified extracts of
existing drugs.
3Data in the PMA Annual Survey Report, 1987-1989, indicate that 82.9% of member lirm
company-financed human and veterinary use R&D expenditures in 1987 were spent on research
for the advancement of scientific knowledge and development of new products and related
services. The remainder was spent on research oriented to signiIicant improvements and/or
modification of existing products. This percentage has not varied much over the last few
decades. The reason why such a high proportion of R&D is spent on new drugs is that a
compound patent can usually be obtained covering the active ingredient. This patent confers
protection from competition from any generic version of the drug.
&The two most recent rate of return analyses in the economics literature use cost estimates in
Hansen (1979. 1980). Joplekar and Paterson (1985) estimates an after-tax rate of return for a
hypothetical NCE f;rst investigated in 1976. ‘Gradowski and Vernon (1990) probides rate of
return estimates for NCEs first marketed in the U.S. in the 1970s. The cost estimates used are
applicable to NCEs lirst investigated in humans during the period 1963-1975. Future studies can
benefit from cost estimates relevant to a later period.
sWithin the past few years, there have been several major mergers and acquisitions in the
pharmaceutical industry including SmithKline-Beecham, Bristol Myers-Squibb, Eastman KodakSterling Drug, Merrell Dow-Marion
Laboratories,
and American Home ProductsRobins.
‘For studies of competitiveness in pharmaceuticals, see the analyses presented in the National
Academy of Engineering (1983), Grabowski (1989). and especially Thomas (1990).
‘In particular, Congressional concern has been focused on increases in relative drug prices
and the role of increased R&D costs in explaining higher prices [U.S. Congress (1985, 1987,
1989)].
J.A. DiMasi et al., Cost of innocation in pharmaceutical industry
109
The ouput from our work can be applied to these various issues in
subsequent research.
The approach to estimating an average pre-tax cost of new drug development used in this paper follows, for the most part, that found in Hansen
(1979). Micro-level data on the cost and timing of development were
obtained through a confidential survey of U.S. pharmaceutical firms for a
random sample of NCEs first investigated in humans from 1970 through
1982. Reported development expenditures run through 1987. For every NCE
that is approved, several others are abandoned at some point in the
development process. Consequently, we associate the costs of failed projects
with those of successful projects. R&D is also treated here as an investment
with expenditures capitalized to the point of marketing approval.
In this paper, we examine the sensitivity of R&D costs to various
parameters such as the clinical success rate and the economic discount rate.
We also consider the sensitivity of R&D costs in reductions in Food and
Drug Administration (FDA) regulatory review times and the lengths of
various phases of NCE development. In addition, results are presented for
the costs of developing approved new drugs categorized by an FDA measure
of a new drug’s medical significance.
The remainder of this paper is organized as follows: section 2 contains an
outline of the new drug development process, which serves as background for
the rest of the paper; section 3 provides a review of the literature on the cost
of drug development; the cost estimation procedures we employ are described in section 4; a description of our sample data and the population from
which they were drawn is given in section 5; baseline results on the average
cost of NCE development are presented in section 6, while results for
subsamples are given in section 7; section 8 contains an analysis using
published aggregate pharmaceutical industry R&D data that serves as an
external check on some of our baseline results; and some conclusions and
prospects for future research are offered in section 9.
2. The new drug development process
New drug development is typically a sequential process. At several points
in the process a pharmaceutical firm will review the status of testing on a
drug and make a decision on whether to continue with its development.* In
general, the decision depends on potential therapeutic benefits, the expected
frequency and severity of adverse reactions, projected additional development, marketing, distribution, and production costs and estimates of a future
revenue stream.
sThe number and timing of these critical points in the life of an investigational drug vary by
firm. See Wiggins (1981) for a discussion of when these decisions are usually made and what
considerations are used in making them.
J.A. DiMasi et al., Cost of innovation in pharmaceutical industry
110
For drugs that make it all the way through to the point of FDA
marketing approval, the sequence usually runs as follows. Prior to synthesis
considerable discovery research is undertaken by chemists and biologists to
develop concepts for new compounds. Once a new compound has been
synthesized it will be screened for pharmacologic activity and toxicity in
vitro, and then in animals. If, at this point, the drug is still considered a
promising candidate for further development, the firm will file with the FDA
an Investigational New Drug Application (IND).9 Unless the FDA places a
hold on this application, the firm may begin clinical (human) testing 30 days
after filing. Clinical testing normally occurs over three distinct phases, each of
which contributes different amounts and types of information on safety and
efficacy.
Phase I testing is performed on a small number of (usually healthy)
volunteers. These trials are conducted mainly to obtain information on
toxicity and safe dosing ranges in humans. Data are also gathered on the
drug’s absorption and distribution in the body, its metabolic effects, and the
rate and manner in which the drug is eliminated from the body.
In the second phase of human testing, phase II, the drug is administered to
a larger number of individuals. The groups selected consist of patients for
whom the drug is intended to be of benefit. Under the 1962 Amendments to
the Food, Drug, and Cosmetics Act of 1938 (FD&C), substantial evidence of
efficacy in the intended use of the drug is required before marketing approval
can be granted. When successful, phase II trials usually provide the first
significant evidence of efficacy. Additional safety data are also obtained
during this phase.
The third, and final, premarketing clinical development phase, phase III,
involves large-scale trials on patients. Additional evidence of efficacy is
sought during this phase. The larger sample sizes increase the likelihood that
actual benefits will be found to be statistically significant. Because many
patients are typically enrolled in the trials, phase III testing is also useful in
detecting adverse reactions that occur infrequently in patient populations. In
addition, this testing may more closely approximate the manner in which the
drug would be utilized after marketing approval.
Although extensive toxicology experimentation on animals occurs during
the preclinical period, to detect teratologic and carcinogenic effects fnms
usually perform long-term animal testing concurrent with phases II and III.
Long-term stability testing, and sometimes additional dosage formulation
work and process development for manufacturing the compound in sufficient
quantities for clinical testing, also occurs during the clinical period.
Once the clinical development phases have been completed and the firm
believes that it has sufficient evidence for approval, it will submit a New
?ke
Mattison
et al. (1988) for analysis of trends in the number of INDs filed by U.S. firms.
J.A. DiMasi et
al., Cost of innovation in pharmaceutical industry
111
Drug Application (NDA) to the FDA for review. Marketing for approved
uses may begin upon notification from the FDA.”
We designed our procedures for estimating average costs to utilize this
categorization of drug development by phase. Not all new drug testing fits
neatly into this framework, but the firms were able to allocate clinical costs
satisfactorily to phases.
3. Previous estimates of new drug R&D costs
A number of studies, covering various time periods, have attempted to
provide estimates of at least a portion of the R&D expenditures required to
bring new drugs to market. Although comparing results from studies with
different methodologies is risky, taken as a whole these analyses point to
rising real R&D costs over time.
Table 1 presents the main results from these studies, the nature of the data
analyzed, and the periods covered. Schnee (1974), Sarett (1974), and Clymer
(1970) used out-of-pocket development cost data on limited groups of NCEs.
Clymer’s estimates for the late 196Os, exclusive of the costs of unsuccessful
products, range from $2.5 to $4.5 million. This may be compared to Schnee’s
estimate of $0.5 million. If both estimates are correct, then R&D costs
increased dramatically from the 1950s to the late 1960s.”
Mund (1970) and Baily (1972) used annual U.S. industry R&D expenditures and NCE introductions to estimate average costs, and also found a
substantial increase in costs over time. An advantege to using aggregate data
is that estimates will be based on information from a large group of firms. A
serious disadvantage, though, is that aggregate expenditures cannot be
associated in any precise manner with particular NCEs.”
Figure 1 shows the time pattern between 1963 and 1989 of annual U.S.
pharmaceutical industry R&D expenditures and U.S. NCE approvals. As
this figure shows, NCE approvals have changed only moderately over time,
while real dollar pharmaceutical outlays have increased several times over
this 25 year period. Aggregative industry data are therefore strongly suggestive of sharp increases in R&D cost over time, and this upward trend seems
to have intensified in the 1980s. There are a number of difficulties, however,
‘OThis contrasts with subsequent FDA approval of additional indications, dosage strengths,
dosage forms, and delivery systems for the new drug. Further testing is required to obtain
approval for such changes in the drug’s labehing.
“Sarett’s estimate for 1967 is also consistent with the Clyrner range. A plausible hypothesis
for explaining at least part of the large observed post-1962 change is that more extensive testing
was required for approval following the implementation of the 1962 Amendments to the FD&C.
‘*Mund utilizes a simple Z.-year fixed lag approach. Baily relates the average of aggregate
R&D expenditures lagged 4 to 6 years to NCE introductions using a regression analysis
framework. In computing the Baily cost estimates in table 1. we employ actual values of his
independent variables, as opposed to the ‘steady state’ equilibrium value used in his article. The
former is more appropriate for comparative purposes.
Aggregate industry data
on R&D and NCES approved
197% 19x5 (by therapculic class)
Wiggins
Includes discovery and development costs
capitalized to date of marketing introduction
Fixed lag regression analysis approach; also uses
selective parameter estimates from Hansen
125 mil(1986 dollars)
cost studies
54 mil(1976 dollars)
Full
Representative sample of drugs tested in
humans, 196331975
Hansen
(B)
Fixed lag regression analysis approach;
no capitalization
2.3 mil/late 1950s
2 I .8mil/late 1960s
Aggregate industry
data on R&D and NCEs
Baily
Discovery costs and unsuccessful NCEs ignored;
no capitalization
Five-year lag assumed between R&D and new
drug introductions; no capitalization
Mund
1.2 mil/1962
3.0 mil/1967
Il.5 mi1/1972
Discovery costs and unsuccessful NCEs ignored;
no capitalization
I .5 mil/l95Os
10_20mil/1960s
New drug candidates
Single drug Rrm
Clvmer
cost studies
0.5 mil/l950s-1960s
Partial
Aggregate industry
data on R&D and NCEs
Successful NCEs
Single drug firm
Sarret
(A)
Notes
Discovery costs ignored; no capitalization
17 approved NCEs
Single drug tirm
Schnee
R&D cost/period
10.5 mil/late 1960s
Sample
Author
Table 1
Prior R&D cost studies.
J.A. DiMasi
RID (BILLIONS
a--
-REAL
et al.,
Cost of
innovation
in
pharmaceutical industry
I
OF 1980 0
113
OF NCEa
40
R&D
6-
-30
4-
- 20
2-
- 10
0
I
83
85
67
69
71
73
I
I
75
71
YEAR
I
I
1
1
1
79
81
03
35
97
-0
89
Fig. 1. Annual U.S. pharmaceutical industry real R&D expenditures and U.S. NCE approvals
for the period 1963-1989. Expenditures are for PMA member firms and are indexed using the
GNP Implicit Price Deflator. The NCE approval line is constructed from 3-year moving
averages of annual approvals.
in using aggregate data to obtain precise estimates of R&D cost per
approved NCE over particular subperiods. First, one must specify a lag
structure between R&D inputs and output. This lag has been lengthening
over time. Second, aggregate industry data cover compounds licensed from
abroad, in addition to those that are U.S.-discovered. These two categories of
drugs are likely to have vastly diferent R&D cost expenditure patterns
within the United States. i3 For these reasons, R&D cost studies based on
micro data at the individual project level are likely to produce much more
accurate cost estimates.
The only prior R&D cost study that utilized multi-firm project level data
for new drugs is Hansen (1979). Data were obtained from 14 firms on a large
randomly selected group of NCEs first tested in humans during the period
1963 to 1975. Information was requested from surveyed fums on NCEs
originated and developed by those firms. Time and cost data by development
phase were gathered so that average phase lengths and costs could be
computed. Costs of projects that were abandoned at some point in the
clinical testing period were included and R&D was treated as an investment
l3For example, the aggregate industry R&D cost data available from the PMA do not
include the R&D expenditures performed abroad for licensed compounds originating in foreign
countries. As we show later in the paper, where aggregate data are employed as a check on our
results, average U.S. R&D costs for foreign licensed compounds are estimated to be small
compared to the self-originated NCEs of U.S. firms. Of course, license fees account for much of
the difference in R&D costs, but they are not publicly available.
114
J.A. DiMasi et al., Cost of innovation in pharmaceutical industry
with returns delayed until marketing approval. Hansen’s R&D cost estimate
of S54 million (in 1976 dollars) was larger than those of prior studies, but it
was also more complete in the sense that it took account of discovery as well
as development costs, the costs of failed projects, and the time value of funds
invested in R&D.
The most recent drug R&D cost study is Wiggins (1987). It employs
industry aggregates calculated at the therapeutic class level to compute an
‘R&D production function’. Regression coefficient estimates are then used to
infer an R&D cost value. This methodology is similar to that employed in
Baily (1972). For NCEs approved during 1970-1985, Wiggins found an
uncapitalized cost per NCE of $65 million in 1986 dollars. Wiggins then used
the time profile for new drug development estimated in Hansen (1979) to
provide capitalized cost estimates. For his base case, Wiggins used Hansen’s
preferred discount rate of 8% to determine capitalized cost to be $125 million
in 1986 dollars. Woltman (1989), however, has pointed out that the Wiggins
capitalized value should only be $108 million, given a proper analysis of the
time profile utilized in Hansen (1979).i4
When compared to Hansen’s results, the corrected estimates reveal a
relatively small real cost increase. General price inflation accounts for 80% of
the increase in estimates ($54 million in 1976 dollars translates to $97 million
in 1986 dollars). However, the time periods considered in the two studies also
do not differ very much. Wiggins used data that were meant to be relevant
to 1970-1985 NCE approvals. Given his imposed lag structure, R&D
expenditures included run from 1965 through 1982. Hansen’s R&D expenditures run from 1963 to the mid-1970s, with expenditures more concentrated
in the latter half of this period.15
Even for identical periods, however, results from these two types of studies
may not be comparable. Wiggins employed data that are a mixture of both
licensed and self-originated NCEs, and this is likely to produce a lower
R&D cost estimate than from a sample consisting of only U.S. selforiginated NCEs. i6 In addition, Wiggins’ (and Baily’s) estimates, which are
obtained from an R&D production
function type analysis, are really
marginal costs. They are estimates of the additional cost that would have
been incurred if one more NCE had been approved during the sample
r4To obtain capitalized costs, Wiggins multiplied his 565 million uncapitalized cost estimate
by the ratio of capitalized to uncapitalized costs in Hansen (1979). Uncapitalized costs per
marketed NCE are not presented in Hansen (1979). but can be determined from other data
therein. Wiggins claimed that this figure is $28 million. The correct amount, however, is S32.5
million. This changes the Wiggins base amount to $108 million.
ISHansen’s data are for a sample of NCEs first investigated in humans during 1963-1975.
Some of the NCEs first investigated in the early part of this period had long development times
with significant expenditures incurred late in the period.
r6As noted in footnote 13, licensed compounds originating from foreign countries tend to
have much lower U.S. costs than compounds originating in the U.S. from U.S.-owned firms.
J.A. DiMasi et al., Cost of innovation in pharmaceutical industry
115
period. Hansen’s results, and those of the other studies listed in table 1, are
average cost estimates. Unless constant returns to scale with respect to R&D
effort exist at the industry level (at or near average expenditure levels for the
sample period), marginal and average costs differ.
In summary, results from the studies discussed above suggest that new
drug R&D costs have risen over time, at least relative to the pre-1962
Amendments period. However, only Hansen’s study employs micro data, and
none of the studies includes much data from the 198Os, a period during
which real industry R&D expenditures have risen rapidly (fig. l).” This
study, like Hansen’s, employs project level data, and a significant amount of
the data used was taken from this recent time period.
4. Estimating the average cost of new drug development
We use project level data to obtain an estimate of the clinical period
average cost per NCE tested in humans. A majority of these NCEs will not
be approved for marketing. Our approach uses an estimated clinical
approval success rate to link the R&D costs of failed projects to the costs of
those NCEs that obtain marketing approval. A representative time profile for
an NCE progressing through all testing phases to the point of marketing
approval is also estimated and used to capitalize R&D expenditures.
Preclinical discovery and development costs are accounted for and treated in
a more refined manner than was the case in Hansen (1979).
Since the full R&D costs for licensed or acquired NCEs are not typically
reflected in R&D budgets of the firms that acquired them, we restricted our
analysis to self-originated NCEs. The various steps we use to determine an
estimate of the total cost per approved self-originated NCE (inclusive of
unsuccessful efforts, preclinical expenditures, and the opportunity cost of
funds invested) are formally described below.
4.1. Expected out-of-pocket costs
Let h be the clinical period development cost of a randomly selected NCE
tested in humans. This can be decomposed into a sum of random variables.
Specifically, h =x, + x,, +x,,, + xA, where x,, x,, and xl,, represent the NCE’s
development costs for phases I, II, and III, respectively, and x* is its
long-term animal testing cost. If the project was terminated prior to entering
a phase, then the random variable for that phase assumes the value zero. The
expected value of clinical period costs is then E(h) =p,pclIc + p,,plllc +p,,,pIIIIe +
“Data from the PMA Annual Survey Report, 1987-1989, can be used to compute growth
rates for real pharmaceutical R&D expenditures for different time periods. From 1963 to 1979
real R&D expenditures grew at a 5.3% compound annual rate; from 1980 to 1989 expenditures
grew at a lO.T/, compound annual rate.
116
J.A. DiMasi et al., Cost of innovation
in pharmaceutical
industry
and PHI are the probabilities that a randomly selected
NCE tested in humans will enter phases I, II, and III, respectively, pA is the
probability that long-term animal testing will be done, and ~ller p,,,_ ~l,,,e,
and pAle are conditional expectations. SpeCifiCally,
,uLllc, pIlIe, p,lllc,
and pAlc
are the population mean costs for phases I, II, III, and long-term animal
testing, respectively, for those NCEs that enter the respective phase,
The proportion of NCEs tested in humans that are taken through the
various clinical period testing phases diminishes with each successive phase.
As will be shown below, only a small minority undergo all testing phases.
Thus, a simple random sample of this population will likely contain a good
deal of information about NCEs that do not last very long in testing and
little information about NCEs that reach the point of NDA submission or
approval. Furthermore, we expect phase costs to be more variable for later
testing phases. The later clinical trials are larger, last longer, and should be
more subject to variability in size and duration depending on the type of
drug tested and the condition it is meant to treat.
Therefore, to reduce overall sampling error, we grouped NCEs into strata
according to the time spent in active testing if research was abandoned,
whether research was still in progress, and whether an NDA had been
submitted or approved. Successful NCEs were deliberately oversampled and
unsuccessful NCEs lasting only a short time in active testing were undersampled. The sample percentage distribution for the strata was specified
before sample selection. In preparing our estimates, we reweighted the
responses to replicate the population.
The population from which our sample was selected consisted of a subset
of the investigational NCEs contained in a Center for the Study of Drug
Development (CSDD) database. The proportions of population NCEs that
fall in our strata can be determined from information in this database. Thus,
we were able to define a set of weights to be applied to the sample data that
transform the sample from one that is unrepresentative with respect to the
strata to one that is perfectly representative. Mean phase cost (in constant
dollars), Xi, and the probability that an NCE tested in humans will enter a
phase, si, can then be estimated from the weighted sample (see appendices A
and B for computational details). The expected clinical period cost per new
drug tested in humans can be expressed as E =s,Z, +s,&, +s,,,i,,, +s,z?,.
To estimate the expected cost per approved drug, C,, one needs to
multiply i? by the ratio of the number of drugs taken into humans, nT, to the
number approved, n, (i.e., C U= i?(nr/n,)). Alternatively, C, can be expressed
as g//s, where s is the successful rate for NCEs tested in humans (the ratio of
n, to nT). Data from CSDD databases on investigational and approved
NCEs for the survey firms can be used to estimate s.
The final step in estimating R&D costs is to incorporate preclinical
expenditures. Firms are generally not able to allocate all preclinical costs to
PAPAWS,wherePI, PII,
J.A. DiMasi et al., Cost of innovation in pharmaceutical
industry
117
ACTUAL
PHASE
LENGTHS
ADJUSTED
PHASE
LENGTHS
Fig. 2. Representative
time profile for an NCE passing
through
all clinical
development
phases.
specific NCEs in a precise manner. Consequently, we use reported aggregate
annual firm R&D expenditure data for the preclinical and clinical periods,
and our previously estimated cost per approved NCE estimate, C,, to derive
P,, the uncapitalized preclinical cost per approved NCE. Specifically, we use
aggregate data to estimate the ratio of preclinical period to total expenditures, 1. We then estimate P, from the identity ,l-P,/(C,
+P,), or P, G
[A/( 1 - J)]C,. Estimated total uncapitalized cost per approved NCE is given
as C,+P,.
4.2. Expected capitalized
costs
To include the opportunity cost of funds invested in NCE R&D for a full
cost estimate, we capitalize the mean phase costs to the point of marketing
approval using a representative time profile for an NCE passing through all
clinical period testing phases; use of a representative time profile yields a
lower bound on costs. Weighted mean phase lengths, ti, are calculated for
our survey NCEs in the same manner as weighted mean phase costs.
It is generally not the case that a phase will begin exactly when the
preceding phase ends. In many instances there is a significant overlap for
successive phases. There can also be gaps between phases. To construct a
time profile we reduce each calculated mean phase length by the weighted
average overlap of that phase with the next (or add to the phase length for a
gap). l8 Figure 2 dep icts a hypothetical time profile. The adjusted mean
phase lengths, cr, are used to determine where in the time profile phase costs
begin. Hence, the time from the start of phase I to NDA approval, T, is
given as T = cl+ I;;+ tfi + tN, where tR is average NDA review time. We
measure the time from the start of a phase to the expected approval date, Zj,
as zj=~~~~ rr, with t,=O.
Mean long-term animal testing time, tA, is calculated in a manner identical
“For
phase III the succeeding
phase is the period
during
which the NDA is under
review.
J.A. DiMasi et al., Cost of innovation in pharmaceutical industry
118
to that used for the actual mean durations of the clinical trial phases. The
length of this phase, however, is not adjusted for an overlap (or gap) with
other phases. We need to determine, though, how this testing tits into our
representative time profile. The weighted mean time from the start of phase I
to the start of long-term animal testing, t,,, can be calculated. We then
consider long-term animal testing costs to begin t,, units of time after the
start of phase I.
Phase costs are assumed to be distributed uniformly over actual mean
phase lengths and capitalized at the discount rate r. Specifically, capitalized
mean phase costs, cj, are determined as follows, where the index values 1, 2,
and 3 refer to phases I, II, and III, respectively.
cj=
y (Zj/tj)e”dt
=I-tl
forj=
1,2,3.
(1)
Capitalized long-term animal testing costs, cA, can be found in a similar
fashion.lg
An estimate of expected capitalized clinical period cost per NCE tested in
humans, ?, can be obtained as 2 = s,c, + s,,c,, + sl,,clll+ sAcA. Transforming this
into a cost per marketed NCE, C,, requires only the approval success rate.
In particular, we have C, = Z//s.
Eq. (1) implies continuous compounding. The discount rate estimates we
use are based on discrete compounding. It is possible, though, to find a
continuous rate, r, that is equivalent to any discrete rate (with regard to
monetary value after the same amount of time).20
Preclinical costs can be capitalized to the point of marketing approval in
the same way that phase costs are in eq. (1). The average duration of the
preclinical period, t,, can be estimated using data on preclinical development
times from a CSDD database on approved NCEs. Capitalized preclinical
cost per approved NCE, P,, can then be given as
T+r,,
P,
=
i
(P&J
e” dt.
(2)
4.3. Cost of capital for pharmaceutical
R&D
In selecting a baseline cost of capital
‘%peciIkally,
capitalized
long-term
for this analysis, we sought
animal testing costs can be computed
1::
cA =
J &/I~) e”dt,
where
ty = T-t,,
and
ti = t)f- t,.
*!z
*‘For example, if the relevant discrete rate is lo”/
we use r=0.095.
as
a
J.A. DiMasi et al., Cost of innovation
in pharmaceutical industry
119
representative value for the pharmaceutical industry over the period spanned
by the development of the drugs in our study. Our sample includes NCEs
first tested in humans between 1970 and 1982. The marketing approval dates
for this sample centers around the mid 198Os, with a few compounds still in
active research. Hence, we want a cost of capital measure that essentially
covers the period between 1975 and 1985 where the bulk of the expenditures
are concentrated.
Our baseline analysis employs a 9% real cost of capital. This value is
based on a recent study on R&D returns [Grabowski and Vernon (1990)].
The study applied the capital asset pricing model (CAPM) to a portfolio for
a representative sample of pharmaceutical firms for each of the years from
the mid 1970s to the mid 1980s. Over this period, the equity cost of capital
for this representative group of firms clusters around 9%.‘r The capital
structure of the pharmaceutical industry is overwhelmingly equity financed
(in excess of 90%). Hence, their estimated 9% cost of equity capital offers a
good proxy for the overall cost of capital.22
Hansen (1979) utilized an 8% cost of capital. That study focused on an
earlier time period (i.e. R&D expenditures in the 1960s and 1970s). Although
the 8% value was not based explicitly on a CAPM analysis, it is generally
consistent with the values emerging from analyses for the relevant period of
that study [see, e.g., Statman (1983)]. The slightly higher cost of capital
utilized in the present study essentially reflects the trend toward higher real
interest rates during the past decade [Feldstein (1988)]. Although a 9% real
cost of capital measure is employed as our preferred estimate in the baseline
analysis, we also perform sensitivity analysis using values significantly above
and below this value.23
5. Data
Twelve U.S.-owned pharmaceutical
firms agreed to participate in this
study and provide, on a confidential basis, information on their self*‘An analysis of investment
riskiness for this portfolio of pharmaceutical
firms was undertaken and indicated that pharmaceutical
firms generally had comparable
riskiness to the market
over the period (i.e. betas approximately
equal to one). Long-term
estimates on risk-free rates
and market rates were obtained
from Ibbotson
Associates
[Grabowski
and Vernon (1990, p.
8’Wl.
220ne can compute a weighted cost of capital, taking account of how much debt relative to
equity is held by pharmaceutical
firms, as well as the tax deductibility
of debt compared
to
equity [see Statman (1983)]. Since the percentage
of capital held in the form of debt is minimal
over this period, this does not change the computed
9% value in an appreciable
manner.
*‘An alternative
approach
for estimating
the cost of capital would be to utilize the arbitrage
pricing model (APT). Some preliminary
analysis
applying
the APT to the pharmaceutical
industry by one of the authors suggests higher values for the cost of capital than that obtained
from the CAPM. The hurdle rates employed
by pharmaceutical
firms also tend to suggest
somewhat higher values for the cost of capital [Grabowski
and Vernon (1990)].
120
J.A. DiMasi et al., Cost of innovation
in pharmaceutical industry
originated NCE development activities (a copy of the questionnaire is
available upon request). 24 These firms include a number of the largest in the
U.S. pharmaceutical industry, as well as some that are small in size. From
1980 to 1986, these firms accounted for approximately 40% of U.S. industry
R&D expenditures on ethical pharmaceuticals.25
We limited our study to self-originated NCEs since the full research and
development expenditures for licensed and acquired NCEs would not be
reflected in the acquiring firm’s R&D accounts. The NCEs included in the
study were first tested in humans during the period 1970-1982. Most of the
NCEs from this period that eventually emerge as approved drugs will have
entered the market in the 1980s or early 1990s. We did not extend the
sample past 1982 since many of the later NCEs would still be in early testing
at the time of the survey.
To construct a representative sample of NCEs tested during this period we
used information from a CSDD database on investigational NCEs. This
database, obtained from a triennial survey of the U.S. pharmaceutical
industry,26 provided us with information on the origin, therapeutic area, and
development status of NCEs first tested in humans during the period 19701982. This information allowed us to identify the survey firm NCEs that met
our study inclusion criteria (i.e. self-originated NCEs first tested in humans
during the period 1970-1982), and to select a random sample of survey firm
NCES.~’ Detailed information was requested from each firm for the NCEs
that we selected.
The firms provided data on 93 NCEs, or 19% of all NCEs in the CSDD
database that met inclusion criteria. The NCE therapeutic class distributions
for the sample and population are nearly identical. For the NCEs selected,
firms provided clinical period phase beginning and ending dates, costs by
year of clinical testing, and the date testing was suspended on those NCEs
for which research was abandoned.
Aggregate pharmaceutical R&D expenditure data for the years 1970-1986
were also provided by the survey firms. The firms reported total annual
pharmaceutical R&D expenditures broken down into expenditures on seiforiginated NCEs, on licensed or otherwise acquired compounds, and on
existing approved products. For firms as a whole during the period 197024A number of these firms had several major pharmaceutical
divisions. Counting
these as
separate units, data were obtained from 15 divisions.
“One of the survey firms was not able to break out pharmaceutical
R&D from total R&D
expenditures
for the 1970s. The industry
ievel is the aggregate
of PMA member firm real
pharmaceutical
R&D expenditures
over the period 1980-1986 (computed
from data in PMA
Annual Survey Report, 1987-1989).
26At the time the sample was selected, 43 firms were included in the database,
34 of them
U.S.-owned.
“The eligible NCEs were first stratified as described in section 4, and a random sample was
selected from each strata.
J.A. DiMasi
et al., Cost of innovation in pharmaceutical industry
I21
Table 2
Average out-of-pocket
clinical period costs for NCEs tested in humans
thousands of 1987 dollars).’
Standard
deviation
of phase
costs
Testing
phase
Mean
cost
Phase I
Phase II
Phase III
Long-term animal
Other animal
2,134 4,519
3,954
5,230
12,801 13,974
2,155 2,411
648
1,183
Total
Median
cost
Nb
Probability
of entering
phase (%)
960
2,175
7,888
1,336
129
87
70
36
49
15
100.0
75.0
36.2
56.1
15.8
(in
Expected
cost
2,134
2,966
4,634
1,209
102
11,045
‘All costs were deflated using the GNP Implicit Price Dellator. Weighted values
were used in calculating means, standard deviations, medians and the probability
of entering a phase.
“N = number of NCEs with full cost data for the phase.
1986, 73.7x, 16.3x, and 10.0% of total pharmaceutical R&D expenditures
were spent on self-originated NCEs, existing approved products, and licensed
and acquired compounds, respectively. Annual total expenditures for selforiginated NCE R&D were further broken down according to whether they
were incurred during the preclinical or clinical periods (since firms are
generally unable to allocate all preclinical costs to individual NCEs, data
were not obtained on preclinical expenditures for specific NCEs).
The firms did not provide full phase cost data for every NCE that entered
a given phase, either because that phase was ongoing on the survey date or
records were incomplete. Costs were not included in computations if the
phase was ongoing or cost data were missing. Only nine NCEs were
continuing in research on the survey date, and for each of them complete
information is available for some phases.
6. Baseline
Clinical
per investigational
Using the
procedure described
section 4,
phase costs
calculated and
presented in
2. All
were deflated
the
GNP
Price Deflator
are given
1987 dollars.**
a
“The
Ideally, we
index,
than the
Implicit Price
measures changes
the prices
the nation’s
want to
an index
on pharmaceutical
input prices.
such
exists. Other
which we
were judged
be less
Implicit Price
The Producer
Index is
solely by
output prices.
(1987) develops
R&D input
index applicable
a number
industries for
years 1969-1983.
closest match
the pharmaceutical
is the
and oil
Only a
pharmaceutical lirms,
were
122
J.A. DiMasi et al., Cost of innovation
in pharmaceutical
industry
number of firms reported clinical period animal testing other than the
long-term animal tests, we reported their costs as other animal costs. On
average, however, they constitute a very small portion of total costs.
The phase to phase attrition rates in column 6 of table 2 are roughly
consistent with alternative estimates in an FDA study [Tucker et al. (1988)].
That study used FDA records to follow the progress of the new molecular
entities (NMEs)~’ for which INDs were filed during the years 1976-1978.
Based on their records and a mathematical model to project future outcomes, they found 70% of the NMEs entering phase II and 33% entering
phase III.
Average costs vary substantially by clinical development phase. Phase II
costs are nearly twice as large as phase I costs. The large-scale clinical trial
costs of phase III are about six times as large as phase I costs. These are
averages, however, and the third column of table 2 reveals substantial
variation in individual NCE testing costs for each phase. The phase cost
distributions are also positively skewed, as indicated by the differences
between mean and median costs.
Since there is substantial attrition of NCEs prior to the more expensive
phases II and III, expected phase costs rise less steeply than mean costs. For
example, expected phase III costs are only slightly more than twice as large
as expected phase I costs. Our estimate of the expected clinical period cost
for NCEs tested in humans is $11.0 million in 1987 dollars.
6.2. Capitalized clinical cost per investigational NCE
To calculate the capitalized value of clinical period expenditures, we
computed adjusted phase lengths and constructed a representative develop(footnote 28 continued)
included in the sample. We deemed the chemicals and oil index too broad-based to use.
Extrapolation of the index through 1987 would also have been necessary. We also considered an
index developed for biomedical R&D [U.S. Bureau of Economic Analysis (1986)], but judged it
to be too dependent on the university research sector. However, we did apply alternative price
indices to our cost data and found the following to be the case in general. The Producer Price
Index resulted in overall cost estimates that were roughly
- . 10% lower than those obtained with
the GNP Deflator. We regressed Mansfield’s chemicals and on index on the GNP Deflator to
obtain nredicted Mansfield index values for 1983-1987. Annlying the expanded Mansfield index
resulted in overall costs that were approximately 10% higher- than the GNP Deflator estimates.
Use of the biomedical R&D index resulted in about a 10% increase in costs relative to those for
the Mansfield index.
29The FDA definition of an NME differs somewhat from our definition of an NCE. One
difference is that the FDA includes diagnostic agents and we do not. Another difference is that
we include a few therapeutically significant biologics, whereas the FDA does not include any.
Aside from the fact that we study a longer time period, an additional reason why the FDA study
results are not precisely comparable to ours is that their drugs include some that were licensed
or acquired. Presumably these drugs were prescreened to some extent, and so we might expect
them to have success rates higher than those for self-originated new drugs. The impact of the
other differences, however, is unclear.
J.A. DiMasi
et al., Cost of innovation
in pharmaceutical
industry
123
Table 3
Average
Testing
phase
lengths
phase
Phase I
Phase II
Phase III
Long-term animal
Other animal
and
clinical period
thousands
capitalized
costs
of 1987 dollars).’
for NCEs
tested
Mean phase
length
(actual)b
Mean phase
length
(adjusted)b
Time from
phase start
to approval’
Capitalized
mean
phase costd
15.5
24.3
36.0
33.6
33.6
16.2
22.5
29.9
33.6
33.6
98.9
82.7
60.2
78.7
78.7
4,103
6,564
17,370
3,366
1,012
Total
in humans
(in
Capitalized
expected
phase cost
4,103
4,924
6,288
1,889
159
17,363
‘All costs were deflated using GNP Implicit Price Deflator.
bMean phase lengths and the time from the beginning of a phase to NDA approval are given
in months. Weighted values were used in calculating
mean phase lengths.
‘The NDA review period was estimated to last 30.3 months. Animal testing was estimated to
start 4.0 months into phase II.
%osts were capitalized at a 9% discount rate.
ment time profile as described in section 4.2. The actual and adjusted phase
lengths presented in table 3 indicate, on average, a very small gap (0.7
months) between the end of phase I and the beginning of phase II, and an
overlap of phase III with both phase II (1.8 months) and the NDA review
phase (6.1 months). The mean NDA review time for self-originated NCEs
first tested in humans during 1970-1982 by U.S.-owned firms (30.3 months)
was determined from a CSDD database. Long-term animal testing began, on
average, 20.2 months after phase I testing commenced.30 Our estimated time
profile has a length of 98.9 months from the initiation of clinical testing to
NDA approval. The time from the start of phase I testing to NDA
submission is 68.6 months.
The expected capitalized cost per NCE tested in humans for the entire
clinical period is $17.3 million in 1987 dollars. Of that amount, 36% is
accounted for by interest costs.
6.3. Clinical costs per approved NCE
To relate costs to the number of approved NCEs, we need a clinical period
approval success rate estimate. While phase success rates could only be
obtained from information in our cost survey, approval success rate estimates
can be obtained using a broader sample contained in CSDD databases on
investigational and approved NCEs. Our methodology for estimating the
“‘Other animal testing sometimes occurred before long-term animal testing, sometimes after it,
and at still other times both before and after it. No tendency in relation to the timing of longterm animal testing was apparent.
As an approximation,
then, we distributed
other animal costs
over the long-term animal testing period.
124
J.A. DiMasi et al., Cost of innovation in pharmaceutical industry
approval success rate is similar to that used in Sheck et al. (1984) and is
described in appendix B. Evidence that the predictive power of our model is
quite high is presented in the appendix.
The base clinical success rate for the cost survey firms (23%) is utilized to
determine clinical cost per approved NCE from our prior estimates of clinical
cost per investigational NCE presented in tables 2 and 3. In particular,
dividing our prior cost estimates by the success rate of 23% yields estimates
of clinical costs per approved NCE. Utilizing a 9% discount rate, the
capitalized clinical cost per approved NCE is estimated to be $75 million,
while the uncapitalized cost per approved NCE is $48 million. This contrasts
with estimates of $17.3 and $11 million for capitalized and uncapitalized cost
per investigational NCE, respectively.
6.4. Preclinical costs
In the preceeding sections we reported expected clinical period development costs calculated from project specific data. Many costs incurred during
the preclinical period cannot be directly assigned to specific NCEs. Moreover, some specific development projects are abandoned prior to reaching the
clinical stage. The activities in the preclinical period are essential to the
eventual development of NCEs and the expenditures during this period
should be allocated to approved NCEs to determine the total cost of
developing a new pharmaceutical.
The activities during the preclinical period lead to a flow of NCEs into
clinical testing. As described in section 4.1, we use the ratio of uncapitalized
preclinical to clinical expenditures and our estimate of the uncapitalized
clinical period cost per approved NCE to allocate preclinical costs to
approved NCEs. In order to estimate A, the preclinical period portion of
total R&D costs, we need to confine our analysis to those expenditures that
involve self-originated NCEs. Therefore, in our survey we requested annual
expenditures on self-originated NCE R&D by preclinical and clinical periods
for each of the survey firms for the years 1970-1986.
In the aggregate, 66.1% of total self-originated NCE R&D was spent
during the preclinical period. Both the preclinical and clinical expenditure
series, however, tend to increase in real terms over the survey period. Since
clinical period expenditures lag behind preclinical costs, the ratio of preclinical period real R&D expenditures to total real R&D expenditures overestimates the true preclinical period contribution to the total. A lag structure for
the aggregate data, based on average lengths of the preclinical and clinical
periods, can be imposed to better estimate II.
The mean length of the preclinical period for self-originated NCEs of U.S.owned tirms first tested in humans during 197G-1982 was estimated from
data in a CSDD database on approved NCEs to be 42.6 months. Our
J.A. DiMasi
et al., Cost of innovation
in pharmaceutical
industry
125
Table 4
Expected
Testing
phase costs per marketed
NCE (in millions
Uncapitalized
expected cost
phaseb
Preclinical
Phase I
Phase II
Phase III
Long-term animal
Other animal
65.5
9.3
12.9
20.2
5.3
0.4
Total
113.6
of 1987 dollars).’
Mean phase length
(actual)
Capitalized
expected costC
42.6
15.5
24.3
36.0
33.6
33.6
155.6
17.8
21.4
27.1
8.2
0.7
230.8
‘All costs were deflated using the GNP Implicit Price Deflator. A 23% clinical
approval success rate was utilized.
bathe NDA review period was estimated to last 30.3 months. Animal testing was
estimated to start 4.0 months into phase II.
CCosts were capitalized at a 9% discount rate.
representative time profile suggests a clinical period duration (inclusive of
phase III testing after NDA submission) of 74.7 months. We approximate the
lag between preclinical and clinical period aggregate expenditures by computing the time between the midpoint of the preclinical period to the
midpoint of the clinical period. 3’ Thus, our base case lag is 5 years and the
corresponding value of 1 is 0.577. Hence, our previously estimated uncapitalized clinical cost is multiplied by a factor of 1.36 to obtain an estimate of
uncapitalized preclinical cost (P, s [A/( 1 - n)]C,).
Using the 5 year lag, uncapitalized and capitalized preclinical costs are $66
and $156 million, respectively. For lags of 4 and 6 years (adjusting at the
same time the assumed duration of the preclinical period), the uncapitalized
costs are $69 and $63 million and the capitalized costs are $152 and $166
million, respectively. Thus, our preclinical cost estimates are not very
sensitive to reasonable variation in the lag.
6.5.
expected
cost
approved
the preferred 5 year lag,
23%, total capitalized cost
million.
research
in
entail
particular, interest costs represent
year from
months)
data
phase
start of
NCE
discount rate
9x, and clinical success
per approved NCE
shown in
4 as
activities, because they occur much
interst costs than
clinical activities. In
58%
preclinical costs ($90
of
costs were provided by year
clinical testing. Computing
costs by
I testing revealed that
midway
of the clinical period
to
time of
expected cost
NCE tested
humans.
126
J.A. DiMasi
et al., Cost of innooation
in pharmaceutical
industry
Table 5
Preclinical, clinical, and total capitalized expected costs per marketed
NCE at alternative approval success rates and discount rates (in
millions of 1987 dollars).’
Capitalized costs
0%
5%
8%
9%
10%
15%
(A) Success rate: 25%
Preclinical
Clinical
Total
61
44
105
98
57
155
I31
66
197
144
69
213
156
73
229
247
93
340
(B) Success rate: 23%
Preclinical
Clinical
Total
66
48
114
107
62
169
142
72
214
156
75
231
170
79
249
269
101
370
(C) Success rate: 20%
Preclinical
Clinical
Total
76
55
132
123
71
194
163
83
246
179
86
265
196
91
287
309
116
425
‘All costs were deflated using the GNP Implicit Price Deflator.
Clinical period expenditures were assumed to represent 42.3% of
total R&D expenditures. The corresponding assumed preclinical
phase length is 42.6 months.
$156 million), whereas they represent only 36% of clinical costs ($27 million
of $75 million).
Table 5 allows us to examine how sensitive our estimate of total R&D
cost per approved NCE is to variations in two key parameters - the clinical
success rate and the discount rate. If we take as plausible ranges on these
parameters a discount rate of 8 to 10% and a clinical success rate of 20 to
25x, the corresponding range in R&D cost per approved NCE is between
$197 and $287 million. This provides some bounds around our estimate of
$231 million per approved NCE by allowing for uncertainty in the values of
these important parameters.
6.6. Comparison
with Hansen’s
analysis
It is interesting to compare our base case average cost estimates with the
preferred estimates in Hansen (1979), and measure the contribution to the
apparent increase in costs of changes in several factors. In 1987 dollars
Hansen’s total capitalized cost estimate is $100.5 million, while his clinical
period capitalized cost estimate is $39.2 million, Thus, in real terms, total
capitalized costs are about 2.3 times larger here than in Hansen; clinical
J.A. DiMasi et al., Cost of innovation
in pharmaceutical
industry
127
period capitalized costs are about 1.9 times larger. Assuming both sets of
estimates are correct for their respective covered time periods, we can
examine how much of the cost increases are due to differences in a given
determinant of cost, holding all other determinants fixed.
Ignoring second-order interaction effects we may state the following. Table
5 reveals that the higher opportunity cost of capital used here (9% as
opposed to 8%) results in an increase in cost amounting to 13.0% of the
measured increase in total cost between the two studies.
R&D and regulatory review times are somewhat longer here. Our results
show essentially no difference in phase II duration, but there is a 7.1 month
increase in phase I duration and a 7.3 month increase for phase III.
Regulatory review time and the preclinical period are 6.3 and 6.6 months
longer here, respectively. Our longer development and regulatory review
times result in total cost increases amounting to 23.9% of the measured
increase in total costs.
The effects on total costs of changes in the discount rate and phase
development times, taken together, constitute the change in time costs
between the two studies. The residual proportion of the measured increase in
total costs (63.1%) can be viewed as that part of the change in measured
total costs that is due to changes in out-of-pocket costs. Out-of-pocket costs
are dependent on the clinical approval success rate, the probabilities of
entering the various phases, and the mean uncapitalized phase costs.
Mean uncapitalized preclinical period and clinical phase costs are substantially higher here and account for the bulk of the increase in out-of-pocket
costs. In constant dollars, phases I, II, and III mean uncapitalized costs are
3.9, 1.4, and 2.5 times larger here than in the Hansen study. Similarly, longterm animal and preclinical costs are 1.6 and 3.7 times larger here,
respectively. The phase-to-phase success rates estimated here are higher than
those reported by Hansen, 32 but their cost increase effect is more than offset
by the cost reducing effect of a higher clinical approval success rate.
4.7. Reductions
in development
and review times
Since our estimates of tbe full cost of developing a new pharmaceutical
include an implicit interest charge, changes in the timing of expenditures will
affect the full cost. Firms may be able to shorten development times by doing
parallel, rather than sequential, studies. Clinical trial phase lengths may also
3ZFewer drugs are screened out in phases
probability
of reaching phases II and III are
the Hansen study, respectively. The probability
increased from 25% to 31%. but the approval
remained virtually unchanged
(65.8% previously
I and II here relative to the earlier study. The
75% and 36% in this study and 50% and 19% in
of approval conditional
on reaching phase II has
probability
conditional
on entering phase III has
vs. 63.5% currently).
128
J.A. DiMasi et al., Cost of innovation
in pharmaceutical industry
Table 6
Impact on capitalized expected costs per marketed NCE of
one-year reductions in average phase lengths (in millions of
1987 dollars).’
One-year reduction
Total capitalized
expected cost”
Reduction in
total cost
Preclinical
Phase I
Phase II
Phase III
NDA review
224
218
215
213
212
8
13
16
18
19
“All costs were deflated using the GNP Implicit Price
Deflator. A 23% clinical approval success rate was utilized.
‘Costs were capitalized at a 9% discount rate.
be shortened if the conferences now held between the FDA and sponsor
firms during the development period provide firms with a better understanding of what will be necessary for approval. Reductions in the average length
of the NDA review phase would shorten the time between R&D expenditures and marketing approval and so lower total costs.
In table 6 we present estimates of the effects of a l-year reduction in phase
duration (holding mean phase costs constant). The largest cost reduction
($19 million or 8.2% of total cost) is for a l-year reduction in the NDA
review phase. The reason the NDA phase carries the greatest impact is that a
reduction in any phase length decreases the capitalized cost for that phase
and for all prior phases.
Since preclinical costs are a large portion of the total cost, even a l-year
reduction in phase 1 produces savings of $13.5 million per approved NCE.
Approval rates during the 1980s have averaged 19 per year. Thus, our results
suggest that l-year reductions in the clinical development or regulatory
review phases would, holding all else constant, produce annual savings of
$257 to $361 million in 1987 dollars.33
The activities required to shorten development times may require additional out-of-pocket costs. For example, doing studies in parallel may result
in performing tests that would not have been performed had the results of
the other tests been known. Shortening the NDA review period may require
additional resources for the FDA. The cost of achieving a reduction in
development times must be compared to the estimated savings in implicit
interest costs. Within a broader context, one should also consider the effects
on product lifetimes and associated sales revenues from having earlier
marketing dates, and the benefits to society from having effective medications
available for patient use sooner.
‘aIt is worth noting that the entire FDA budget for fiscal year 1988 was S478 million, with
$169 million spent on salaries and expenses for the human drugs and biologics divisions [U.S.
Department of Health and Human Services (1990)].
J.A. DiMasi et al., Cost of innovation
in pharmaceutical industry
129
7. Cost estimates for subsamples
7.1. Within-sample period effects
We examined whether it is possible to discern time trends in our data. The
sample did not contain enough information to investigate costs on a year-toyear basis, so we divided the sample into two time periods (NCEs first tested
in humans during the periods 1970-1976 and 1977-1982). The distribution of
population NCEs over the strata used in the weighting scheme for the full
sample differed by period. Consequently, we reweighted NCEs for each
period according to the sample and population strata proportions for that
period.
Mean phase III development time and cost are significantly lower for the
more recent period, but these data are not complete. In particular, NCEs
with long development times will be underrepresented in the late period. At
this point, meaningful comparisons across periods of phase III costs cannot
be undertaken. However, counting only phase I and phase II costs, the 19771982 period uncapitalized costs are 100.5% higher than those for the 19701976 period. This suggests a strong upward time trend in out-of-pocket
clinical costs.
7.2. Clinical period costs for approved NCEs
The cost of development may differ between successful and unsuccessful
projects. To investigate this, we examined costs for the approved NCEs in
our sample. We had complete cost data on 22 of the 27 NCEs that had
obtained marketing approval as of October 31, 1990. One of the 22 NCEs is
an extreme outlier with respect to cost and was removed from the approved
NCE subsample.
It was retained for the full sample results. Given the
much larger sample size and the small weight this NCE received in the
weighting scheme used for the full sample, its impact on the full sample
results is minor.
Table 7 offers a comparison between mean phase costs and lengths for the
approved NCEs and those for all sample NCEs. Development times are very
similar. However, except for early clinical testing (phase I), phase costs are
significantly higher for the approved NCEs. This may reflect a tendency to
direct more resources, perhaps through conducting more studies concurrently, to the NCEs that early clinical testing suggests have the greatest
promise of approval. Table 7 also shows that, as is the case for the full
“Both the uncapitalized and capitalized clinical costs of the outlier are roughly double those
of the approved NCE with the next highest clinical costs.
130
J.A. DiMosi et al., Cost of innovation in pharmaceutical industry
Table 7
Clinical period average phase costs and phase lengths for approved NCEs and for all sample
NCEs (in thousands of 1987 dollars).’
Approved NCEsb
Full sample’
Testing
phase
Mean
phase cost
Standard
deviation
of phase
costs
Phase I
Phase II
Phase III
Long-term
animal
Other animal
2,475
5,629
20,024
2,957
4,138
14,016
14.0
25.9
36.8
2,134
3,954
12,801
4,519
5,230
13,974
15.5
24.3
36.0
3,646
1,777
3,065
2,300
37.4
37.4
2,155
648
2,411
1,183
33.6
33.6
Mean
phase length
(actual)
Mean
phase cost
Standard
deviation
of phase
costs
Mean
phase length
(actual)
‘All costs were deflated using the GNP Implicit Price Deflator.
bEstimates for approved NCEs are based on data for 21 of the 93 sample NCEs.
‘Weighted values were used in calculating means and standard deviations for the full sample
estimates.
sample, substantial variation exists in the costs of individual approved drugs
for each phase.
The uncapitalized mean clinical period cost for the approved NCEs is
$31.9 million; the median cost is $31.0 million. The 95% confidence interval
for mean out-of-pocket clinical period cost is $31.9f7.7 million.35 Since we
have approval dates for these drugs, it was not necessary to use a
representative time profile; costs were capitalized to the point of actual
marketing approval. Mean capitalized clinical period cost for approved
NCEs is $43.0 million; median capitalized cost is $40.9 million. The 95%
confidence interval is $43.0 & 11.3 million.36*37
The FDA rates new drugs according to whether the drug represents an
important therapeutic gain (A), a moderate therapeutic gain (B), or little or
no therapeutic gain (C) over existing therapy. To get a sense for whether the
cost of developing NCEs is related to their therapeutic importance, we
35A chi-squared goodness-of-lit test was used to compare the sample frequency distribution of
uncapitalized clinical period costs to a normal probability distribution. The hypothesis of no
difference in distributions could not be rejected. Hence, we used a t-distribution in constructing
the confidence interval.
36As is the case with uncapitalized costs, the hypothesis that the capitalized costs were taken
from a normal population cannot be rejected.
“‘Given the right-censored nature of the data, approved NCEs with very long development
times may be somewhat under-represented
in the sample. We found a moderate positive
correlation between capitalized clinical period cost and the time from the start of clinical testing
to NDA approval (Spearman rank correlation coeffcient=0.661). Thus, reported costs for the
approved NCEs may be somewhat biased downward.
J.A. DiMasi
et al., Cost of innovation in pharmaceutical industry
131
computed costs for approved NCEs according to their therapeutic ratings.-”
Mean and median capitalized clinical period costs for the A and B rated
approved NCEs are $51.7 and $41.5 million, respectively.” Costs for the C
rated approved NCEs are somewhat lower - mean and median capitalized
clinical period costs are $36.5 and $36.2 million, respectively.
8. An external check on the results
As an external check on our results, we analyzed the R&D performance of
the U.S. pharmaceutical industry to see if it was consistent with our analysis.
In particular, our approach is to relate industry R&D expenditures to the
NCE introductions of U.S. firms (see fig. 1) using the same lag structure and
time period in our study. However, since our study focuses on the selforiginated NCEs of U.S.-owned firms, it is necessary to make various
transformations of the publicly available industry data before meaningful
comparisons can be undertaken.
For the cost survey firms, we determined the proportions
of their
aggregate pharmaceutical
R&D expenditures that were spent on selforiginated NCEs (73.7%) and on licensed or acquired NCEs (10.0%).
Published PMA industry R&D expenditure data, however, do not distinguish between what firms spend on self-originated NCEs and what they
spend on licensed or acquired NCEs. The procedures we employ to estimate
annual industry self-originated NCE R&D expenditures and how they can
be related to annual NCE approvals are described in appendix C.
We averaged both lagged R&D expenditures and the number of approvals
over the period 1979-1989.40 Using an average of 7.3 industry self-originated
approvals per year for this time frame, we found uncapitalized and capitalized cost estimates of $138 and $270 million, respectively. In 1983 there was
only one approval of a self-originated NCE from a U.S.-owned firm.
Excluding this outlier year yields 7.9 as an average annual number of
“Although the FDA gives an initial rating to investigational NCEs shortly after IND tiling,
we have information only on the tinal ratings for the approved NCEs in our sample. The FDA
rating system must be viewed with caution when considered as a measure of therapeutic benefit
since the FDA has rather limited information when it assigns the ratings. The ultimate clinical
significance of some drugs may not become apparent until the drugs have been in widespread
use for a number of years for the original approved indications, or until their effectiveness for
additional indications has been noted in clinical practice.
‘9The A and B rated drugs were combined since there is only one A rated drug in the
subsample. Of the remaining 20 approved NCEs, eight are B rated and 12 are C rated.
“Year-to-year
cost estimates using aggregate data are highly variable because they are
extremely sensitive to the denominator (number of self-originated NCE approvals) used for the
calculations. Thus, to reduce variability in the estimate, we used average values for the
numerator and denominator, as opposed to an average of the annual ratios of lagged
expenditures to current approvals.
132
J.A. DiMasi et al., Cost of innovation
in pharmaceutical
industry
approvals, and reduces the uncapitalized and capitalized cost estimates to
$127 and $250 million, respectively. Hence, our estimate of $231 million is
comparable to what one obtains from a comprehensive aggregate analysis.
9. Conclusions and prospects for future research
We have estimated the average cost of new drug development from a large
sample of self-originated NCEs that were first tested in humans during 19701982. Data were obtained from a confidential survey of 12 U.S.-owned firms.
The average cost of NCE development was estimated to be $231 million in
1987 dollars (with $114 million of that paid out-of-pocket). A representative
time profile for an NCE passing through all phases of development from
synthesis to marketing approval extended nearly 12 years.
Cost estimates are substantially higher here than in previous studies; some
of which, however, did not measure R&D cost fully or adequately. Our
period of analysis is also more recent than the periods used for these other
analyses. In particular, this study is the only one to date that has reflected in
its estimates much of the sharp increase in pharmaceutical R&D expenditures that occurred during the 1980s.
The most recent study on the cost of new drug development [Wiggins
(1987)] used R&D expenditure data that covered the period 1965-1982, but
with full weight placed only on expenditures from 1967 to 1980. This can
partially explain why costs in that study are much lower than those found
here. Another, perhaps more important, reason is that Wiggins used expenditure and approval data for all NCE approvals. In particular, he did not
differentiate between self-originated and licensed or acquired NCEs. The
R&D expenditures incurred on new drugs up to the point of licensing or
sale will not be included in U.S. industry (PMA) data if the drug originated
from an overseas operation of a foreign firm. Our analysis of industry data
revealed substantially lower company expenditures per new drug on licensed
or acquired NCEs than is the case for self-originated NCEs.
When results here are compared to those of a previous study [Hansen
(1979)] with a similar methodology, total cost is seen to have increased 2.3
times in real terms. Development times are longer and a higher discount rate
is used. The bulk of the increase in cost per approved NCE, however, can be
attributed to large increases in out-of-pocket costs.
That clinical trial costs have risen sharply in recent years is attested to in
F-D-C Reports: The Pink Sheet (1989), where an executive of a major
pharmaceutical firm reports that the information required to support NDAs
has increased dramatically. Clinical trials for one of the firm’s anti-infective
NCEs approved in 1979 used 1,493 patients; the trials for a related antiinfective that the firm is currently developing will require testing on 10,000
J.A. DiMasi
et al., Cost of innovation
in pharmaceutical
industry
133
patients. Also mentioned as factors leading to rising costs are the complexity
and scope of the research required and the adoption of expensive new
technologies. Another factor often suggested to explain increased costs is that
firms are now focusing development more on treatments for chronic and
degenerative diseases, which typically require longer and more expensive
testing.
In many applications that utilize pharmaceutical R&D cost estimates, the
relevant concept is the after-tax cost (e.g., rate of return analyses). R&D is
typically expensed immediately, rather than depreciated during the life of the
marketed product. If the firm has sufficient profits against which to expense
the R&D, the net cash flow required to finance the R&D will be reduced by
the amount of the tax savings. However, in contrast to an investment in an
asset such as a building, which can be depreciated over its useful life for tax
purposes, the investment in R&D necessary to bring a product to market
cannot be depreciated during the years in which sales occur. Thus, the effect
of expensing rather than depreciating R&D is one of changing the time at
which the deduction occurs.
The R&D tax credit, enacted in 1981, was in effect for only part of our
study period. Depending on changes in firm spending levels, the credit had
differential impacts across firms. In aggregate, though, the impact of the
credit on our cost estimates is likely to be minor.41 Since we were only
interested here in estimating social cost, rather than rates of return, we did
not adjust for the timing of taxes or the extra credits.42
Further research on the cost of new drug development could profitably be
directed at a detailed analysis of differences in costs across therapeutic
classes. We will address this issue in future research. Another interesting
topic for future research is whether trade-offs exist on the margin between
clinical trial expenditures and NCE failure rates.
Our analysis has been directed to the R&D cost of self-originated NCEs
from U.S.-owned firms. In future research it would be interesting to analyze
the R&D costs of NCEs originating in other countries. This would be
especially the case for Japan. Aggregate data suggests that R&D costs per
411f we use U.S. pharmaceutical
industry
R&D expenditure
data, we can obtain a rough
upper bound estimate of the effect of the R&D tax credit. Until the Tax Reform Act of 1986
was enacted, the annual credit was 25% of the increase in covered R&D expenditures
over the
average of R&D expenditures
for the previous three years. The Tax Reform Act of 1986 reduced
the credit to 20”/,. Nominal U.S. pharmaceutical
industry R&D expenditures
grew at an 11.7%
compound
annual
rate of growth
from 1978 to 1986. This implies a 6.8% subsidy if all
expenditures
are covered and the credit is 25%. Much of our clinical cost data and the bulk of
the related preclinical expenditures,
though, predate the credit. Thus, the tax savings relevant to
our study are likely much less than 6.8% of total expenditures.
4*The impact of the tax rate in rate of return studies may not be substantial.
Grabowski
and
Vernon (1989) found their results little affected when sensitivity analysis was conducted
on the
tax rate parameter.
134
J.A. DiMasi
et al., Cost of innovation
in pharmaceutical
industry
NCE originating in Japan are significantly lower than those in the United
States [Grabowski (1990)]. At the same time, pharmaceutical R&D in Japan
has different characteristics from that undertaken in the United States, and
has produced significantly fewer NCE introductions
in major markets
outside of Japan. Given the increased policy interest in international
competitiveness, the cost and determinants of R&D performed in different
countries is likely to be an important question for future research.
It would also be instructive to compare the total costs for U.S. firms of
originating a new drug in-house versus licensing or acquiring one externally.
Our survey data indicate that the R&D costs incurred by U.S. firms for a
self-originated compound is several times what they spend for a licensed or
acquired NCE. This raises the issue of whether the licensing fees and other
costs for externally acquired compounds exactly offset the higher R&D costs
of self-originated NCEs. This is what would be expected in equilibrium, if the
market for licensed and acquired NCEs is perfectly competitive. Whether this
in fact has been true of recent experience in the pharmaceutical industry is
an interesting topic for future research.
Another important topic for future work is the effect of firm size and the
scale of R&D activities on the costs of new drug introductions. This topic is
also likely to have significant policy interest, given the recent tendency
toward increased consolidation
of the pharmaceutical
industry. In this
regard, it would be useful to know whether this consolidation is motivated in
part by higher real R&D costs or potential R&D scale economies.
Our R&D cost estimates should be a useful input to studies of the returns
to current new drug introductions. It should be emphasized that rising real
R&D costs do not necessarily represent a problem for pharmaceutical
innovative activity. In particular, if the higher costs are reflective of higher
probabilities of commercial success and/or higher average sales per new drug
introduction, then strong incentives to undertake pharmaceutical R&D can
be maintained even in the face of rising R&D costs. Hence, it will be
important in future research to examine trends in innovative output and
sales revenues, as well as R&D costs.
Finally, it should be emphasized that while our estimates are presented in
1987 dollars, they are not estimates of the cost of developing a new drug in
1987. The drugs included in the sample were first tested in humans between
1970 and 1982, and many of the successful candidates from this sample are
entering the market in the 1980s and 1990s (the mean approval date for
approved drugs in the sample was in early 1983, with future approvals
expected to raise the mean to early 1984). However, the expected cost of
developing a new drug for which testing begins in the late 1980s or the 1990s
will be affected by changes in the process which have occurred since our
sample period. The trends suggest that the cost will be higher than our
estimate.
J.A. DiMasi et al., Cost of innooation
in pharmaceutical
industry
135
Appendix A
Estimated weighted mean phase cost, lip is determined as follows (where
the index values 1, 2, 3 and A refer to phases I, II, III and long-term animal
testing, respectively):
n
_fi= i
j=l
wixij
C
wjnij
fori=l,2,3andA,
(A.11
j=l
where
n =number
Wj
=
of NCEs in the sample,
ratio of the % of population NCES to the y0 of sample NCEs
for the stratum in which NCEj falls,
x,,= jth NCE phase i cost
11 0
i
if the jth NCE has phase i costs,
otherwise,
n,, = 1 if the jth NCE has phase i costs,
IJ 0 otherwise.
I
For phases I, II, and III the estimated probability of entering a phase, sir is
the product of the estimated conditional probability that the NCE will
undergo testing in the phase given that the previous phase was entered, and
the estimated probability of entering the preceding phase. For long-term
animal testing, the estimate, sA, is a weighted proportion of the sample NCEs
that underwent that type of testing. Specifically, we have
(A.3
(A.3)
where
O,=ratio of the % of population
stratum g,
NCEs to the % of sample NCEs for
6i,=number
of sample NCEs in stratum g that entered phase i,
6, =number
of sample NCEs in stratum g,
s(J =l,
a,,, 6,.
With (A.l) and (A.2) an estimate of expected clinical period costs for a
randomly selected NCE tested in humans, i?‘, can be expressed as E=
_
_
slxl +s,,x,, + s,,,X,,,+ s,f,. Given that the sample strata are prespecified, the
J.A. DiMasi et al., Cost of innovation in pharmaceutical industry
136
weights, wj and 0, are not random variables. We can see, then, that Zi and si
are unbiased estimates of the population mean phase i cost and the
population proportion of NCEs entering phase i, respectively. There is no
reason to expect that the weighted mean cost for a phase and the weighted
proportion of NCEs entering that phase are correlated across independent
samples. Under the assumption that Xi and si are stochastically independent,
it can easily be shown that i? is an unbiased estimate of E(h).
Appendix B
The problem of estimating an approval success rate for the population of
cost survey firm NCEs that meet survey inclusion criteria is modelled in two
stages. Since the ultimate fate of some NCEs first tested in humans from
1970 to 1982 is, to our knowledge, not yet determined, survival analysis
should prove useful [see Cox and Oakes (1984)].
In the first stage we estimate the distribution of times that the NCEs are in
residence. For NCEs that have either been abandoned or approved by the
end of 1986, we define residence time to be the time from first testing in
humans to either research abandonment or NDA approval. Some NCEs
were continuing in research as of the end of 1986. Data for such NCEs are
right-censored. Survival analysis techniques utilize the information provided
by censored observations.
In the second stage of the model, we estimate the probability that an NCE
will meet a given fate (research abandonment
or NDA approval) as a
function of residence time. Thus, the cumulative probability of success (NDA
approval) as a function of residence time, t, can be given as
s(t) = j f(uV’(4 du,
0
(B.1)
where f(u) is the probability density function for residence time, and P(u) is
the probability that an NCE with a residence of u receives NDA approval at
that time. The density function can be estimated by a nonparametric
product-limit technique.
The data on approvals at various residence times indicate that the process
may be adequately approximated by a sigmoidally shaped distribution. Thus,
we applied the logit and probit models to estimate P(u). Both the logit and
probit models provided good tits and, in fact, were virtually indistinguishable
from one another.43 Computed values of the likelihood ratio index deve43The transformation
probit models showed
rates are also virtually
suggested
in Amemiya (1981) for comparing
coeflkients
in logit and
very small differences in the coefficients. The results on cumulative success
the same when logit and probit specifications
for P(u) are used.
J.A. DiMasi et al., Cost of innovation in pharmaceutical industry
137
loped in McFadden (1974) shows the probit model to have a slightly
superior tit, and so the success rate estimates utilize the probit specification
for P( u).~~
The cost survey firms investigated in humans 279 NCEs that met cost
survey inclusion criteria.45 As of December 31, 1989, 17.2% of these NCEs
had been approved. Our model predicts that approximately 23% of the
population NCEs will have been approved 14 years after first testing in
humans. By the end of 1986 only four NCEs had been in residence longer
than 14 years, and information obtained in the cost survey indicated that
two of these NCEs had been abandoned. We use 23% as our base case
clinical approval success rate.
Evidence of the predictive power of eq. (B.l) can be obtained by focusing
on one subclass of NCEs. Taking note of dates of research abandonment
obtained from the cost survey and approvals after 1986, we were able to
completely characterize the fate of all anti-infective NCEs. The model (using
data through 1986) predicted an approval success rate of 31.7% for this class,
which compares very well with the actual final cumulative success rate of
3 1.0%.
Appendix C
The PMA publishes R&D expenditure data for its member firms (PMA
Annual Survey Report, various years). From these data it is possible to
extract annual expenditures for human use ethical pharmaceuticals. These
expenditures, in real terms for the years 1967-1987, are given in the first
column of table C.l.
At our request, we also obtained information from the PMA on the
percentage of total R&D expenditures accounted for by U.S.-owned member
firms. The PMA had such data only for recent years. We utilized the 19871988 percentage (84.3%) to estimate annual R&D expenditures by U.S.owned firms (column 2 of table C.l). There is some evidence of a slight
decrease in this percentage over time. Given the downward trend in this
percentage, the values in column 2 are likely to be conservative estimates.
We next subtract foreign expenditures from the values of U.S.-owned
R&D expenditures to obtain their domestic R&D expenditures (column 3 of
table C.l). We include only domestic expenditures in our analysis under the
assumption that the foreign expenditures of U.S.-owned firms will be directed
%+ecitically,
we used an iteratively reweighted non-linear least squares routine to estimate
P(u)=F(a+Bu),
where u represents residence time in months and F(.) is the cumulative
distribution
function of the standard normal random variable. The maximum
likelihood
estimates of a and B are -2.3720 and 0.0244, respectively.
4sThe information needed to conduct the survival analysis was available for 276 of these
NCEs.
J.A. DiMasi
138
et al., Cost of innovation in pharmaceutical industry
Table C.l
U.S. pharmaceutical
industry real human use R&D expenditures
(millions of 1987 dollars).’
1967-1987
(1)
(2)
All PMA
(3)
U.S. R&D
of (2)C
(4)
Self-originated
R&D of (3)
972
1,058
1,135
1,203
1,264
1,253
1,247
1,282
1,338
1,367
1,393
1,468
1,501
1,593
1,771
2,047
2,324
2,513
2,774
3,101
3,508
670
729
782
829
871
863
859
883
922
942
960
1,011
1,034
1,098
1,220
1,410
1,601
1,731
1,911
2,137
2,417
Year
lilTIlS
U.S.-owned
PMA fnmsb
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1,363
1,400
1,492
1,583
1,663
1,684
1,785
1,870
1,928
1,989
2.06 1
2,123
2,275
2,549
2,762
3,100
3,462
3,726
4,142
4,703
5,310
1,149
1,180
1,258
1,334
1,402
1,420
1,505
1,576
1,625
1,677
1,737
1,790
1,918
2,149
2,328
2,613
2,918
3,141
3,492
3,965
4,476
‘Expenditures on R&D for ethical pharmaceuticals intended for human use
were obtained from the PMA Annual Survey Report (various years) and
deflated using the GNP Implicit Price Deflator.
bathe expenditures for all tirms in a year were reduced by the average
percentage of member firm R&D expenditures accounted for by U.S.-owned
firms for 1987 and 1988 (source: Gary Persinger, PMA).
‘Computed based on U.S.-owned firm aggregate expenditure levels and data
contained in the PMA Annual Survey Report (various years).
primarily to non-U.S. introductions.
This assumption will also tend to
produce conservative estimates, since at least some of the foreign expenditures of U.S.-owned firms are obviously directed towards approval of new
drugs in the United States and other major world markets.46
The last step in the analysis is to determine the percentage of R&D that is
devoted to self-originated NCEs by U.S.-owned firms. In particular, we need
to separate such R&D expenditures from those on licensed or acquired
NCEs and on R&D for improvements in existing products. It would be
46Although some of the domestic R&D expenditures of U.S. firms may be targeted speciftcally
to products only introduced abroad, our approach will produce a conservative estimate so long
as these outlays are smaller in value than the foreign R&D expenditures of U.S. firms that are
related to domestic introductions. This appears to be a reasonable presumption given the
tendency of U.S. firms to prescreen compounds abroad for the U.S. market.
J.A. DiMasi et al., Cost of innovation in pharmaceutical industry
139
Table C.2
New
chemical entity approvals in the United States, 1979-1989.
Year
Total
NCEs
U.S.-owned PMA
member firm
NCEs
Self-originated
NCEs of U.S.owned PMA firms’
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
13
11
23
22
12
21
26
20
18
16
21
13
9
15
18
6
15
16
13
13
8
10
9
6
6
11
1
8
12
8
10
5
4
‘Information on the origin of compounds was obtained from a
Center for the Study of Drug Development annual survey of firms
with U.S. NCE approvals and supplemented, when necessary, with
published sources.
inappropriate
to assume that the firms who did not participate in our survey
have the same pattern of R&D allocations as our survey firms. NCE output
data show that non-survey U.S.-owned firms are somewhat more reliant on
licensed or acquired NCEs. 47 We assume that the R&D effort necessary to
produce a self-originated NCE relative to a licensed or acquired NCE is the
same for the survey and non-survey firrn~.~~ Using this approach, we
estimate that the non-survey U.S.-owned firms devoted 60.6% of their
pharmaceutical R&D to self-originated NCEs. Given this estimate, and the
fact that our cost survey firms account for 48.3% of the total pharmaceutical
R&D expenditures of all U.S.-owned firms, we can compute a weighted
47For the cost survey firms, 66.7% of their NCE approvals during the period 1979-1989 are
self-originated. For the same period, only 48.1% of the approvals of other U.S.-owned firms are
self-originated.
481n particular, the ratio of expenditures on self-originated NCEs per approval of NCEs of
this type to the expenditures on licensed or acquired NCEs per approval of such NCEs was
estimated to be 3.7 for firms in our survey. In other words, it takes nearly four times the R&D
outlays to produce a self-originated NCE compared to the R&D necessary for a licensed or
acquired NCE for our survey firms (ignoring licensing and acquisition fees). We assume that this
ratio is the same for the U.S.-owned Iirms that are not in our survey sample. We also assume
that the group of foreign-owned PMA member firms spent the same proportion of their U.S.
pharmaceutical R&D on new products as did the survey firms. We then used information
available in the PMA Annual Survey Report (various years) to estimate that non-survey U.S.owned firms spent 78.3% of their pharmaceutical R&D on new products. Using these two
estimates, together with information on the breakdown of NCE approvals between selforiginated and licensed or acquired NCEs for non-survey U.S.-owned firms, we estimated that
these firms spent 60.6% of their pharmaceutical R&D for self-originated NCEs.
140
J.A. DiMasi
et al., Cost of innovation
in pharmaceutical
industry
average estimate of the percentage of pharmaceutical R&D performed by
U.S.-owned firms for self-originated NCEs (66.8%) and apply it to the
previously calculated values in column 3 of table C.l.
We also performed a parallel analysis for the U.S. NCE approvals between
1979 and 1989. These results are presented in table C.2. We first separated
total NCE approvals into those emanating from U.S.-owned PMA member
firms and all other firms. The NCEs from U.S.-owned PMA firms (shown in
column 2) were then further subdivided into self-originated versus licensed or
acquired NCEs. To classify NCEs into these sub-categories, we used the
CSDD annual survey of NCE approvals together with information from
related sources. For the 1979-1989 period, the last column of table C.2
shows the annual number of self-originated NCEs accounted for by U.S.owned PMA member firms. We found that approximately 39% of all U.S.
NCE approvals for this period were for self-originated NCEs of U.S.-owned
PMA firms.
For a lag structure, we use the phase time profile and cost levels found in
this study. This implies that approvals in one year should be associated with
R&D expenditures lagged 2 to 12 years. Monthly expenditures were
calculated and spread over this period according to the phase time proti1e.49
Weights to be attached to each of the lag years can then be determined,50
and used to estimate uncapitalized cost per approval. Given the time profile,
weights for capitalized cost estimates can also be calculated.
“Data from a CSDD database were used to determine that NCEs approved over the last 15
years received their approvals, on average, eight months into the year of approval. This was
used to establish the endpoints of the time profile.
5oIf approval occurs in year t,,, then the-weights for years t-z to t-t2 are 0.038, 0.059, 0.075,
0.083, 0.072, 0.058, 0.105, 0.163, 0.163, 0.163, and 0.020, respectively. These weights represent the
proportions of out-of-pocket expenditures incurred in the years prior to approval.
References
Amemiya, T., 1981, Qualitative response models: A survey, Journal of Economic Literature 19,
1483-1536.
Baily, M.N., 1972, Research and development costs and returns: The U.S. pharmaceutical
industry, Journal of Political Economy 80, no. 1, 7tX85.
Clymer, H., 1970, The changing costs and risks of pharmaceutical innovation, in: J.D. Cooper,
ed., The economics of drug innovation (American University, Washington, DC) 109-124.
Cox, D.R. and D. Oakes, 1984, Analysis of survival data (Chapman and Hall, London).
Feldstein, M., 1988, Halving the pain of budget balance, Wall Street Journal, May 28, 24.
F-DC Reports: The Pink Sheet, 1989, Lilly estimates ‘3,300 person years’ go into development
of successful new molecular entity; current antibiotic project is requiring 10,000 patients, 51.
nos. 51 & 52, 14.
Grabowski, H.G., 1989, An analysis of U.S. international competitiveness in pharmaceuticals,
Managerial and Decision Economics (special issue), Spring, 27-33.
Grabowski, H.G., 1990, Innovation and international competitiveness in pharmaceuticals, in: A.
Heertje and M. Perlman, eds., Evolving technology in market structure: Studies in
Schumpeterian economics (University of Michigan Press, Ann Arbor) 167-185.
J.A. DiMasi
et al., Cost of innovation in pharmaceutical industry
141
Grabowski, H.G. and J.M. Vernon, 1982, A sensitivity analysis of expected profitability of
pharmaceutical research and development, Managerial and Decision Economics 3, no. 1,
36Xl.
Grabowski, H.G. and J.M. Vernon, 1990, A new look at the returns and risks to pharmaceutical
R&D, Management Science 36, no. 7, 804-821.
Grabowski, H.G., J.M. Vernon and L.G. Thomas, 1978, Estimating the effects of regulation on
innovation: An international comparative analysis of the pharmaceutical industry, Journal of
Law and Economics 21, no. 1, 133-163.
Hansen, R.W., 1979, The pharmaceutical development process: Estimates of current development
costs and times and the effects of regulatory changes, in: RI. Chien, ed., Issues in
pharmaceutical economics (Lexington Books, Lexington, MA) 151-187.
Hansen, R.W., 1980, Pharmaceutical development costs by therapeutic categories, University of
Rochester Graduate School of Management Working Paper No. GPB-80-6. March.
Joglekar, P. and M.L. Paterson, 1986, A closer look at the returns and risks of pharmaceutical
R&D, Journal of Health Economics 5, 153-177.
Mansfield, E., 1987, Price indexes for R&D inputs, 1969-1983, Management Science 33, no. 1,
124-129.
Mattison, N., A.G. Trimble and L. Lasagna, 1988, New drug development in the United States,
1963-1984, Clinical Pharmacology and Therapeutics 43, 15-27.
McFadden, D., 1974, Conditional logit analysis of qualitative choice behavior, in: P. Zarembka,
ed., Frontiers in econometrics (Academic Press, New York) 105-142.
Mund, V.F., 1970, The return on investment of the innovative pharmaceutical ftrm, in: J.D.
Cooper, ed., The economics of drug innovation’(American
University, Washington, DC)
125-138.
National Academy of Engineering, 1983, The competitive status of the U.S. pharmaceutical
industry (National Academy Press, Washington, DC).
Pharmaceutical Manufacturers Association, 1987, PMA statistical fact book (PMA, Washington,
DC).
Pharmaceutical Manafacturers Association, Annual survey report, various years (1968-1989)
(PMA, Washington, DC).
Sarett, L.H., 1974, FDA regulations and their influence on future research and development,
Research Management, 18-20.
Schnee, J.E., 1972, Development costs: Determinants and overruns, Journal of Business, 347-374.
Sheck, L., C. Cox, H.T. Davis, A.G. Trimble, W.M. Wardell and R.W. Hansen, 1984, Success
rates in the United States drug development system, Clinical Pharmacology and Therapeutics 36, 574-583.
Statman, M., 1983, Competition in the pharmaceutical industry (American Enterprise Institute,
Studies in Health Policy, Washington, DC).
Thomas, L.G., 1990, Spare the rod and spoil the industry: Vigorous competition and vigorous
regulation promote global competitive advantage, presented at the International Joseph
Schumpeter Society Meetings, June, Charlottesville. VA.
Tucker, S.A., C. Bloxan and P. Coppinger, 1988, The outcome of research on new molecular
entities commencing clinical research in the years 197678, Office of Planning and
Evaluation Study 77 (Food and Drug Administration, Rockville, MD).
U.S. Bureau of Economic Analysis, Government Division, 1986, The biomedical research and
development price index: Revised and updated estimates (U.S. Government Printing Office,
Washington, DC).
U.S. Congress, House, Subcommittee on Health and the Environment of the Committee on
Energy and Commerce, 1985, Prescription drug price increases, 99th Congress, First Session,
July 15, 1985 (U.S. Government Printing Oflice, Washington, DC).
U.S. Congress, House, Subcommittee on Health and the Environment of the Committee on
Energy and Commerce, 1987, Medical devices and drug issues, 100th Congress, First Session,
April 27, 1987 and May 4, 1987 (U.S. Government Printing Oflice, Washington, DC).
U.S. Congress, Senate, Special Committee on Aging, 1989, Prescription drug prices: Are we
getting our money’s worth?, Majority staff report (U.S. Government Printing Office,
Washington, DC).
142
J.A. DiMasi
et al., Cost of innovotion in pharmaceutical industry
U.S. Department of Health and Human Services, 1990, Justilication of appropriation estimates
for committee on appropriations, Food and Drug Administration, Vol. 10 (U.S. Public
Heahh Service, Hyattsville, MD).
Wiggins, S.N., 1981, The pharmaceutical research and development decision process, in: R.B.
Helms, ed., Drugs and health: Issues and policy objectives (American Enterprise Institute,
Washington, DC).
Wiggins, S.N.. 1987, The cost of developing a new drug (Pharmaceutical Manufacturers
Association, Washington, DC), June.
Woltman, H.R., 1989, Reviewing the bidding: R&D costs and profitability of new chemical
entities, Journal of Research in Pharmaceutical Economics 1, no. 3, 49-65.